gflownet.policy.base

Classes

ModelBase	Helper class that provides a standard way to create an ABC using
Policy	Helper class that provides a standard way to create an ABC using

Module Contents

class gflownet.policy.base.ModelBase(config, env, device, float_precision, base=None)

Bases: abc.ABC

Helper class that provides a standard way to create an ABC using inheritance.

device

float

state_dim

fixed_output

random_output

output_dim

base = None

parse_config(config)

abstract instantiate()

__call__(states)

make_mlp(activation)

Defines an MLP with no top layer activation If share_weight == True,

baseModel (the model with which weights are to be shared) must be provided

Parameters:

• layers_dim (list) – Dimensionality of each layer

• activation (Activation) – Activation function

class gflownet.policy.base.Policy(config, env, device, float_precision, base=None)

Bases: ModelBase

Helper class that provides a standard way to create an ABC using inheritance.

instantiate()

fixed_distribution(states)

Returns the fixed distribution specified by the environment.

Parameters:

states (tensor) – The states for which the fixed distribution is to be returned

random_distribution(states)

Returns the random distribution specified by the environment.

Parameters:

states (tensor) – The states for which the random distribution is to be returned

uniform_distribution(states)

Return action logits (log probabilities) from a uniform distribution

Parameters:

states (tensor) – The states for which the uniform distribution is to be returned